2x2列聯表之穩健診斷

Abstract

當分析2×2 列聯表時，可利用格內的機率（及勝算比）作為評估的參數；其方法可假 設此列聯表服從一多項分配。當資料存在離群值時，最大概似估計的結果多不具一致 性；因此本計畫之目的，即在建立一k 個2×2 列聯表分析的穩健估計。考慮在許多2× 2 列聯表資料中存在離群值（列聯表）時，藉由最大削減概似估計方法以獲致此多項分 配的格內的機率穩健估計。研究結果將以電腦模擬分析及文獻中實際的資料例子分析 來陳現。

While analyzing 2×2 contingency tables, the probabilities of the four cells (and the odds ratio) are often the important parameters to be investigated. The standard way of estimation and testing is to assume a four-cell multinomial model for a 2×2 table. In the presence of outliers, inferences based on the maximum likelihood estimation for categorical data analysis are generally inconsistent. The purpose of this project is to develop a robust method for analyzing k 2×2 tables. Here we describe a maximum trimmed likelihood estimation technique to estimate the cell probabilities for a multinomial model when some possible outliers exist among several 2×2 tables. The proposed procedure is illustrated using simulation study and some real data sets.